Csillámtrace - my new graphics engine

[Geri]

512x256 ~ 10 billion polygon -> 14 fps (idle)
Athlon2 x4 635 @ 3.3 ghz

[lion]

Is there public demo available?

[Geri]

you can try the two game i posted before.
(however, those are compiled with older versions)

[lion]

I mean lib+headers+sample or viewer which can load some popular format (.obj for example) to try own scene.
PS: have you try use sse\avx to process packet of rays at once? Also you can use Intel\AMD OpenCL (CPU) and achieve sse\avx packet processing using serial C code.

[Geri]

no, the engine itself not available separately.

i am not interested in inline assembly, and i dont want opencl or any other gpgpu related craps around my code at all. thats why i made this engine, to avoid using graphics cards.

[lion]

I've read about GPU, I mean CPU OpenCL, it's easy way to see what you can get if you use C (not C++) code, so you can use same serial C code for your initial CPU path and OpenCL CPU path. Also take a look at this http://embree.github.io/ it's CPU only, have serial C++ code as well as optimized sse\avx.

[Geri]

well this is a very very interesting question, and here is my opinion about this:

-i dont want such dependencies in my code

-i dont really think that opencl would compile better code than g++. of course, maybe in special cases, where opencl is meant to, it would, but in real-life softwares, i think its impossible

-i dont really have any vectorizable code, except some very few lines in non-so-speed-critical positions that would mean nothing to speed-up

-i dont understand, why they force vectorization, in real-world programs, only very very few can benefit from them, since we alreday have superscalar cpus. vector instructions just waste of the transistors

it would maybe an interesting topic, to see, how cpu-s run these gpu-style algorythms, only for scientifical reasons, but i dont really see any point using it in real-world programs.

judging from the ray/sec rate on your link, on an i7 they only have 7 million triangle/sec render time, so very very time-waste conception - jbikker's arauna alreday can do 3x more in real-world in real environment. so this opencl conception for cpu-s is alreday a dead idea.

i would not waste my time on such things, also, i dont really have opportunity to start sculpting my almost 40.000 code long engine to use opencl in the middle of the renderer for such tests, escpecially when the whole point of this engine was to avoid using such things at all. so its not just so easy. i dont see point in this at all.

[lion]

There no opencl on link.
7 million triangle/sec - I think you look at wrong place, we look for rays\s, you need to calculate and show it in one of tutorialXX_ispc samples.
Also note that arauna have sse optimizations and process packet of rays at once, so you need compare it with _ispc version of embree binary. In opposite embree is serial C++ code (like your engine) which turned to packet using ispc compiler.

PS: darktower game is fun, keep that way =)
PS2: better add game links to first post, so all can easily find it.

[Geri]

yes, sorry, i meant ray/sec.

dark tower and the chess game was just basically public beta-tests of the technology, i have learnt a lot of them how NOT to make games in this engine

so i not really linking them. the first heavyweight software maybe arrive with this engine in 2015 Q1, that will be properly advertised on the necessary places.

[Geri]

please, fill this survey. results will be public

http://esurv.org/online-survey.php?surv ... H_73dddd5a